In order to gain a better understanding of the genetic architecture of CMI, we conducted a whole genome linkage screen using a collection of 66 nonsyndromic families with at least two sampled individuals presenting with CMI with or without syringomyelia. It was hypothesized that the limited evidence for linkage across all 66 families collectively was due to genetic heterogeneity and may be associated with the phenotypic variability observed. Based on the co-occurrence of CMI and CTDs, families were stratified by CTD related conditions in order to identify phenotypically and potentially genetically more homogeneous groups of families for linkage analysis. Stratified analyses identified multiple genomic regions showing increased evidence for linkage consistent with reduced genetic heterogeneity across families as a result of the CTD related stratification criteria. Furthermore, several plausible disease genes were identified as discussed in detail below.
Prior to describing our most significant results, it is important to relate our findings to the only other whole genome linkage screen conducted to date which implicated regions on chromosomes 9 and 15 
. We only identified suggestive evidence for linkage to the region on chromosome 9 within our CTD-positive group of families. Importantly, 12/66 of our total families and 7/34 CTD-positive families overlap with the families used in the initial screen conducted by Boyles and colleagues; therefore, these results do not provide independent replication for this region. Lack of replication for chromosome 15 could be due to the use of: 1) different genotyping chips (Illumina Human610-Quad BeadChips versus Affymetrix 10K SNP Chip) and marker quality control procedures, 2) different linkage software packages (Merlin versus Allegro; e.g. different with respect to an error detection option and accounting for inter-marker LD) and genetic models (penetrance function and S scoring function), 3) additional families which are likely genetically heterogeneous, and/or 4) different analytical approaches (stratified analyses). While the original finding could be a false positive, it is equally possible that as additional families are collected and other approaches to reduce genetic heterogeneity are applied to the data this region may present again as a promising candidate genomic interval warranting follow-up.
While we presented linkage results within the subsets of both CTD-positive and CTD-negative families, the focus of the current paper has been on the CTD-negative families as these are thought to represent more “classical” CMI due to cranial constriction and also resulted in the identification of the only genomic region with a maximum LOD score exceeding 3. The most significant of our findings implicated the growth differentiation factors, GDF6 and GDF3, both of which had been previously implicated in KFS 
which is characterized by cervical vertebral fusion and may be associated with a wide range of conditions including renal abnormalities, cardiovascular abnormalities, orthopedic anomalies, pulmonary problems, deafness, and synkinesia 
. Interestingly, roughly 3–5% of CMI patients are diagnosed with KFS 
, suggesting a shared genetic etiology between these disorders. Further, it has been proposed that KFS and CMI should be classified as post-otic neural crest syndromes, thus sharing a common cellular etiology 
. Although the exact relationship between these disorders is unknown, one possibility is that CMI and KFS may be allelic disorders. In order to investigate this possibility, GDF3 and GDF6 were sequenced in a collection of CMI patients from our linkage families. While GDF3 still presents as an intriguing biological candidate and additional sequencing of potential regulatory elements may yield putative disease variants, no variants of obvious significance were identified in this study. However, several interesting variants were identified in GDF6. A previously identified KFS mutation, A249E (rs121909352), was found in two of our CMI families. The functional effect of this mutation has been determined previously in-vitro
. Asai-Coakwell and colleagues evaluated changes to bone morphogenetic protein (BMP) signaling by co-transfecting an expression construct with the A249E mutation and a Sex determining region Y-box 9 (SOX-9)-responsive reporter gene into primary limb mesenchymal cells and assessed SOX-9 reporter activity 
. Reduced activation of the reporter was observed (p<0.034), suggesting altered chondrogenic potential 
. In addition, a 23% reduction in secreted mature GDF6 protein expression was observed for the mutant as determined by Western blot analysis 
Although there is evidence for a functional effect, the expression of A249E is complex with previous evidence of pleiotropy (ocular versus skeletal phenotypes), variable expressivity (e.g. coloboma versus microphthalmia), and reduced penetrance 
. Consistent with these reports, we also observe variable expressivity within our CMI families (CMI with syringomyelia versus CM0). In fact, the identification of A249E in both CMI and a suspected CM0 individual within the same family (9453) further supports the hypothesis that these disorders share an underlying genetic basis and represent part of a continuum of Chiari phenotypes 
. Although A249E is not necessary to cause disease in either of these families, it still likely contributes to disease presentation together with additional genetic and potentially environmental factors.
Additional variants of interest from our study include two intronic GDF6 variants, rs140757891 and a novel SNP, g.406+2780C>T. ChIP-seq data from a small number of cell lines indicate that both variants are located within predicted targets of SUZ12, a polycomb protein involved in epigenetic silencing of developmental genes. Interestingly, haploinsufficient SUZ12 mice exhibit cerebellar herniation, as well as spina bifida, an enlarged brainstem, and occipital cortical changes 
. Although these clinical features appear to be due to an enlarged tectum and only demonstrate partial clinical similarity with CMI, Miro and colleagues suggest that an additional link between SUZ12 and central nervous system disorders may come from neurofibromatosis 1 (NF1), a disorder characterized by the development of neurofibromas and the presence of café-au-lait spots 
. SUZ12 and NF1 are located within 560 kb of each other on chromosome 17 and while most NF1 patients have point mutations in NF1 some harbor larger genomic deletions that encompass NF1 as well as other genes, including SUZ12 resulting in a more severe clinical presentation 
. Roughly 5% of CMI patients present with NF1 
and it has been previously suggested that these two disorders may share an underlying genetic basis 
. Remarkably, within the same group of families that showed increased evidence for linkage to the region containing GDF6 (CTD-negative) we also observed suggestive evidence for linkage to 17p12–q11.2 (Max LOD
2.37, CW emp p-val
0.03) which contains both SUZ12 and NF1 providing further support for a potential role in disease development.
While encouraged by our findings, we acknowledge several limitations of this study. First, because we enforced strict eligibility criteria (exclusion of syndromic cases) and required families to have multiple affected individuals, the total number of families eligible for the study was low and likely contributed to reduced power. However, despite the relatively small sample size, the number of families examined was almost three times as large as the collection of families used in the only other whole genome linkage screen published to date 
. Second, MRIs were not available for all study participants thus misclassification of affection status cannot be ruled out. Importantly, none of our analyses used phenotype information from “unaffected” family members (i.e. affecteds-only analysis), thus the greatest impact of potential misclassification would be if individuals were incorrectly classified as affected. Furthermore, clinical information used for the stratified analysis was mostly ascertained through a general medical interview upon enrollment in the study; therefore, misclassification of families as CTD-positive or CTD-negative is possible. Nevertheless, our data suggest that the increased evidence for linkage observed for the stratified analysis based on CTD related conditions is non-random (e.g. 8q21.3–q22.1: GW emp p-val
0.07, CW emp p-val
0.008). This observation would seem unlikely if a high degree of misclassification existed.
Future work will include functional follow-up of variants of interest as well as sequencing GDF3 and GDF6 in a larger cohort of sporadic and familial CMI cases. Furthermore, the distant regulatory elements previously identified for GDF6 
represent excellent candidate regions for future de novo
variant detection. Other candidate genes, such as low density lipoprotein receptor-related protein 6 (LRP6) present within the chromosome 12 candidate interval could also be investigated as LRP6, when specifically deleted from early mesenchyme, causes a slight delay in mouse skull ossification 
. In addition, rather than simply taking a candidate gene approach, targeted capture and next generation sequencing of candidate genomic intervals defined by linkage analysis or whole genome sequencing would be an obvious next step to comprehensively follow-up these findings. Finally, taking a more quantitative approach to disease, for example by focusing on cranial base morphometrics, may yield greater insight into the genetic etiology due to increased statistical power and reduced misclassification rates among individuals.
The current study demonstrates the utility of using clinical stratification to reduce genetic heterogeneity in CMI by identifying genomic regions showing increased evidence for linkage with maximum LOD scores exceeding 2 and even 3, as well as having implicated credible candidate genes in CMI susceptibility. Although further work is necessary to confirm the involvement of these genes and individual sequence variants in the development of CMI, this work makes several important contributions to the field of CMI research: 1) We conducted the largest whole genome linkage screen to date providing multiple candidate intervals for future investigation and replication, 2) Our results suggest a relationship between CTD related conditions and genetic etiology which is consistent with the hypothesis that CMI with CTDs versus CMI without CTDs occur through different mechanisms (“cranial settling” versus “cranial constriction”), 3) Multiple biological candidates were implicated from the analysis, including the only two GDFs currently known to be associated with KFS suggesting a shared genetic etiology between CMI and KFS. This is consistent with the fact that KFS is known to co-occur with CMI and share a common cellular etiology, 4) Identified a known KFS missense mutation in two of our families that is not necessary for disease but likely contributes to the phenotype due to its rare frequency in the general population, known functional effect in vitro, and the fact that it has been identified in multiple skeletal and ocular disease cohorts, and 5) Identified two potential regulatory variants (one novel, one rare) shared across all affected individuals in the families they were identified in and located within predicted regulatory regions for SUZ12 which itself is an excellent candidate gene for CMI. Further investigation of GDF3 and GDF6, other plausible biological candidates such as SUZ12, NF1, and LRP6, as well as the genetic relationship between CMI and KFS is warranted.